ORBS: language-independent program slicing

Reviewer: Josep Silva

Binkley et al. introduce a novel dynamic slicing technique called observation-based slicing (ORBS). This new technique has a very interesting property: it is language independent, and thus it is able to slice multi-language systems. The underlying idea is to produce slices through repeated statement deletion. After every deletion, the behavior of the original program and the slice is compared. If they produce the same output at some point of interest, then the deletion is accepted. The first implication of this method is that ORBS is useful for those languages where statements can be deleted from the source code. Hence, precision is highly dependent on how the source code is distributed among lines. For instance, two independent statements in a single line cannot be distinguished by this technique due to the lack of a control/data dependence analysis. For instance, code like this: "if ( x >0) { x =42;} else { x =24;}" would either belong to the slice or be removed. But only one branch of the if-statement could not be sliced. Despite the source code limitation, an ORBS slice can eventually be more precise than a dynamic slice because it can delete statements that depend on the slicing criterion (considering the data dependence chain), but that do not really influence it. After introducing ORBS, the authors compare the technique with other similar methods and empirically evaluate their implementation with good results for both precision and performance. In summary, the paper introduces an interesting idea that is properly evaluated and compared. Moreover, it is well written and contains many illustrative examples. Online Computing Reviews Service